面向智能监控的视频摘要与压缩方法研究

发布时间：2018-04-29 07:43

本文选题：监控视频 + 视频摘要　；参考：《合肥工业大学》2014年硕士论文

【摘要】：视频监控系统几乎遍布了城市中的每个角落,不间断地记录着城市发生的一切。监控摄像头为社会安全提供重要保障,同时也带来了海量的视频数据。如何快速从海量视频数据中提取有价值的线索,如何有效压缩存储海量数据,是目前智能视频监控系统亟待解决的问题。本文针对这两个问题,对基于运动分析的视频摘要和视频内容裁剪进行了研究,实现了快速的视频摘要检索和视频内容压缩。本文的主要工作有以下两个方面：第一,对监控视频的摘要进行研究和实现。详细分析了当前主流的视频摘要方法及成熟系统,对其中的共性与关键技术进行了总结。针对监控视频的特殊应用,提出了场景分析的摘要方法。通过分析不同的监控视频产生的场合以及这些场合的固有结构特征,将其分为三类基本场景。针对这三类场景分别设计了三种实用的摘要方法,并用软件实现了一个可用于实际监控视频的摘要系统,该软件已经获批软件著作权。实践表明,此系统能对智能监控系统的视频有效地进行摘要,为快速信息检索奠定了基础。第二,对监控视频的压缩进行研究和实现。针对静态监控视频背景固定、人们只对运动目标区域感兴趣这一实际情况,本文引入图像的线裁剪思想,提出了基于运动感知的监控视频压缩算法。首先通过运动分析方法提取了视频中的ROI区域,然后对非ROI区域进行线裁剪。同时,引入视频帧间的时间和空间一致性约束,保证了视频流的连贯性。实验表明,裁剪压缩后的监控视频在容量上有所减少的同时,最大限度地保留了视频中感兴趣区域目标的信息。
[Abstract]:Video surveillance system has spread around almost every corner of the city, uninterrupted records of what happened in the city. Surveillance cameras provide an important safeguard for social security, but also bring huge amounts of video data. How to quickly extract valuable clues from mass video data and how to compress and store mass data effectively is an urgent problem for intelligent video surveillance system. Aiming at these two problems, this paper studies video summarization and video content clipping based on motion analysis, and realizes fast video summary retrieval and video content compression. The main work of this paper is as follows: first, the summary of surveillance video is researched and implemented. The current mainstream video summary methods and mature systems are analyzed in detail, and the common features and key technologies are summarized. Aiming at the special application of surveillance video, a summary method of scene analysis is proposed. By analyzing the different situations produced by surveillance video and the inherent structural characteristics of these situations, it can be divided into three basic scenarios. For these three scenarios, three kinds of practical methods are designed, and a software is implemented to realize an actual video summary system. The software has been granted the copyright of the software. The practice shows that the system can effectively abstract the video of intelligent surveillance system and lay a foundation for fast information retrieval. Secondly, the compression of surveillance video is studied and implemented. In view of the fact that the background of static surveillance video is fixed and people are only interested in moving target region, this paper introduces the idea of line clipping of image, and proposes a video compression algorithm based on motion perception. Firstly, the ROI region of video is extracted by motion analysis, and then the non-ROI region is clipped. At the same time, time and space consistency constraints between video frames are introduced to ensure the coherence of video streams. The experimental results show that the capacity of the compressed video is reduced and the information of the region of interest in the video is kept as much as possible.
【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN948.6;TP391.41

【相似文献】